EVALUATING FOREST SAMPLING SCHEMES FOR SELECTED NATURAL FOREST TYPES IN KENYA

Abstract

This study determined and compared relative efficiency of commonly used sampling designs, intensities and plot sizes in assessing tropical natural forests. Relative efficiencies of random and systematic sampling designs with four plot sizes (25, 50, 100 and 400 m2) and varying intensities (5, 10, 20 and 30%) were investigated with reference to full-cover 1-ha inventory in selected forest types: tropical rain forest (TRF), moist lower montane forest (MMF) and dry woodland forest (DWF) in Kenya. We hypothesized an ecological gradient from dry woodland foresst through montane to tropical rain forests mirrored by biophysical characteristics; that different forest ecosystems require different sampling protocols; and that optimal sampling designs, intensities and plot sizes exist for different forest attributes. Fifty-six sampling schemes were generated for efficiency testing using computed statistics in R-software and actual field data. Sampling error and effort (time) were intergrated to measure efficiency of each sampling scheme in estimating selected attributes. The three forests formed a complexity gradient in composition, structure, diversity and slope. Different sampling schemes resulted in mixed outcome for tested attributes across the forest types. Efficient schemes for tree regeneration include SSH-10mx5m-5% with 80% efficiency in TRF, SSD-10mx10m-30% with 81% efficiency in MMF and SRS-10mx10m-30% with 90% efficiency in DWF. Schemes for forest density include SSV-20mx20m-20% with 98% efficiency in TRF, SSH-10mx10m-20% with 99% efficiency in MMF and SSV-10mx5m-30% with 92% efficiency in DWF. Quadratic mean diameter sampling schemes are SSH-20mx20m-20% with 94% efficiency in TRF, SRS-10mx10m-20% with 75% efficiency in MMF and SRS-10mx10m-20% with 90% efficiency in DWF. Basal area assessment requires 100% inventory of 1 ha plot, subdivided into 20 m x 20 m data compilation units. For species diversity assessment, most efficient sampling schemes include 30% SSH with any plot size between 5m x 5 m and 20 m x 20 m in TRF and 10% SSH with 10 m x 5 m or 10 m x 10 m plot sizes in DWF. In MMF, all sampling options underestimated actual number of species. Where low sampling intensities are efficient in capturing species richness per hectare, there is no need to spend money and time on higher intensities. To construct cumulative species-area curves, one hectare data recorded in 5 m x 5 m units were found enough across the three forest types. The 30% sampling intensity was significantly favorable for both TRF (SSH-5mx5m-30% and SSH-10mx5m-30%; R2>99%) and MMF (SSH-20mx20m-30%; R2=86%). Slope gradient influenced efficiency of sampling for regeneration and basal area in the montane forest. The suitability of tested sampling options varied among attributes and across forest types. As such, multiple resource inventories require integration of different sampling schemes to ensure efficiency